A group of artificial intelligence and animal ecology specialists has proposed a new, cross-disciplinary strategy to improve wildlife studies and make better use of the massive volumes of data already being gathered owing to modern technologies. Their research was published in Nature Communications today.
Big data and the Internet of Things have invaded the area of animal ecology. Thanks to advanced technology like satellites, drones, and land devices like autonomous cameras and sensors put on animals or in their environs, unprecedented volumes of data on wildlife populations are now being acquired. These data have become so simple to get and transmit that they have reduced distances and time required for researchers while also reducing the disruption of people in natural settings. Today, a range of AI tools for analyzing massive datasets are accessible, but they’re generally broad in character and unsuited to watching the precise behavior and appearance of wild animals. By combining improvements in computer vision with the experience of ecologists, a group of scientists from EPFL and other institutions have described a ground-breaking strategy to solving that challenge and developing more accurate models. Their results, published today in Nature Communications, provide fresh insights into the application of artificial intelligence to aid in the conservation of endangered species.
Developing cross-disciplinary expertise
Wildlife study has progressed from a regional to a worldwide scale. Modern technology now provides ground-breaking new approaches to provide more accurate wildlife population estimates, better understand animal behavior, curb poaching, and reverse biodiversity loss. Using massive datasets, ecologists may utilize AI, notably computer vision, to extract essential elements from pictures, videos, and other visual forms of data to swiftly categorize wildlife species, count individual animals, and glean crucial information. The generic programs that are now used to handle such data typically act as black boxes and do not take use of all of the available information about animals. Furthermore, they are difficult to configure, may suffer from poor quality control, and may be vulnerable to ethical concerns around the usage of sensitive data. They also have biases, particularly geographical biases; for example, if all of the data used to train a software was gathered in Europe, the program may not be acceptable for other parts of the globe.
“We intended to enlist the help of other scholars who were interested in this issue and combine their resources in order to advance this new discipline.” Prof. Devis Tuia, the director of EPFL’s Environmental Computational Science and Earth Observation Laboratory and the study’s primary author, thinks AI “may serve as a crucial catalyst in wildlife research and environmental conservation more widely.” Computer scientists need to be able to rely on the expertise of animal ecologists if they wish to lower the margin of error of an AI algorithm that has been taught to identify a certain species, for example. These experts can specify which characteristics should be considered in the program, such as whether a species can survive at a specific latitude, whether it is important for the survival of another species (for example, through a predator-prey relationship), or whether the species’ physiology changes over time. New machine learning techniques, for example, may be used to automatically identify an animal, such as utilizing a zebra’s distinctive stripe pattern or its movement dynamics in video.” Prof. MackenzieMathis, co-author of the paper and chair of EPFL’s Bertarelli Foundation Chair in Integrative Neuroscience, agrees. “This is where the merging of ecology and machine learning is crucial: the field biologist has extensive domain knowledge about the animal under study, and our task as machine learning researchers is to collaborate with them to develop tools to solve the problem.”
Disseminating information about current projects
Tuia, Mathis, and others addressed their research issues at numerous conferences over the last two years, and the concept of building tighter linkages between computer vision and ecology arose. They recognized that such cooperation may be tremendously beneficial in averting the extinction of some animal species. A number of projects in this area have already been launched; some of these are included in the Nature Communications article. Tuia and his EPFL colleagues, for example, have created a software that can distinguish animal species from drone photographs. It was recently put to the test on a seal colony. Meanwhile, Mathis and her colleagues have released DeepLabCut, an open-source software tool that helps scientists to accurately estimate and monitor animal positions. It has already received 300,000 downloads. DeepLabCut was created for laboratory animals, however it may also be applied for other species. Other colleges have launched programs as well, but it’s difficult for them to share what they’ve learned since there isn’t yet a true community in this field. Other scientists are often unaware that these programs exist, much alone which one would be ideal for their study.
Nonetheless, many internet forums have been used to take the first steps toward forming such a community. The publication in Nature Communications, on the other hand, is aimed at a larger readership of researchers from all around the globe. Tuia states, “A community is slowly taking form.” “So far, we’ve relied on word of mouth to establish a foundation.” We began two years ago with Benjamin Kellenberger, also of EPFL, Sara Beery of Caltech in the United States, and Blair Costelloe of the Max Planck Institute in Germany, who are now the article’s other main authors.”